In what could become a defining moment for AI copyright law, bestselling authors Molly Tanzer and Jennifer Gilmore have filed a class action lawsuit against Salesforce Inc., accusing the tech giant of secretly using thousands of copyrighted books to train its xGen AI models without consent or payment.

Filed in the Northern District of California in October 2025, the case Tanzer et al. v. Salesforce asks one provocative question now echoing across creative and legal circles:

When AI learns from your words, does that count as inspiration, or theft?

The complaint goes far beyond one company’s conduct. It challenges the very foundations of how modern generative AI systems are built, monetized, and justified under the legal shield of “fair use.”

And for Salesforce, a brand long associated with ethical innovation, the optics could be devastating.

How the Lawsuit Unfolded

The 46-page complaint alleges that Salesforce trained its xGen AI models on the so-called Book3 corpus, a massive dataset containing hundreds of thousands of novels, essays, and literary works scraped from the internet, many of them under active copyright.

According to the plaintiffs, these texts were downloaded, stored, and copied in full, forming the linguistic backbone of xGen’s capabilities.

Such acts, they argue, violate the exclusive reproduction rights granted to authors under Section 106 of the U.S. Copyright Act, while giving Salesforce an enormous commercial advantage over creators who received nothing.

Adding to the controversy, the suit highlights public statements by Salesforce’s CEO Marc Benioff, who previously condemned other AI firms for using “stolen data.”

That rhetorical reversal adds a powerful emotional undercurrent and makes this case as much about corporate credibility as copyright law.

The Legal Heart: Fair Use vs. Copyright Protection

To many observers, Tanzer v. Salesforce feels like a sequel to Authors Guild v. Google, the 2015 landmark that allowed Google to digitize books for its search index under the doctrine of transformative fair use.

But the similarities stop there.

What is a copyright lawsuit involving AI?

A copyright lawsuit involving artificial intelligence occurs when creators allege that an AI system used their protected works, such as books, music, or images without permission during model training.

These cases test whether machine learning qualifies as fair use under U.S. law or constitutes unauthorized copying of original content.

Where Google displayed only brief, non-substitutive snippets, Salesforce’s AI training allegedly ingested entire books, creating machine-learning weights that could be used to generate new text in similar style or tone.

The authors claim this process erases the line between study and reproduction, turning human creativity into raw machine fuel.

Salesforce, for its part, is expected to argue that:

• Model training is transformative, producing data representations, not creative copies.

• The process doesn’t compete with the original market, satisfying the fourth fair-use factor.

• Limiting AI training would stifle innovation across industries relying on machine learning.

Understanding what courts mean by “transformative” is key here. As explored in Transformative Fair Use Explained: How to Legally Reuse Works in U.S. Copyright Law, the doctrine allows some reuse, but only when new meaning, message, or purpose is added.

The question now is whether teaching a machine to imitate writing styles qualifies.

Recent rulings such as Court Rules AI Cannot Be Copyrighted: Landmark Ruling on Human Authorship also underscore that copyright demands human input. The Salesforce case now tests the reverse—whether AI can legally consume human works without infringing them.

3. Regulation, Ethics, and the Coming AI Accountability Era

This lawsuit lands amid a broader regulatory awakening. Legislators in Washington are drafting bills that would:

• Require transparency in AI training datasets,

• Create licensing frameworks for copyrighted material, and

• Establish royalty systems compensating creators for data use.

The U.S. Copyright Office is simultaneously reviewing whether AI training qualifies as “reproduction,” potentially setting a new legal threshold for compliance.

If courts act before lawmakers do, Tanzer v. Salesforce could set de facto national policy dictating how AI companies license data in the years ahead.

Salesforce’s case also carries a strong ethical dimension. Benioff’s vocal support for ethical capitalism and responsible tech use may amplify scrutiny.

In an era when investors and consumers value authenticity, perceived hypocrisy in AI ethics could become a reputational liability far greater than the lawsuit’s financial risk.

A Landmark Test for the Future of AI and Copyright

The plaintiffs seek class certification covering thousands of authors whose works were allegedly used in Salesforce’s datasets.

If granted, the financial exposure could reach hundreds of millions of dollars.

Discovery will likely reveal how Salesforce sourced its training data and whether internal discussions acknowledged copyright risks.

Beyond Salesforce, this lawsuit tests whether AI model training equals copying under U.S. law. A plaintiff victory could force developers to license creative content, spawning a new ecosystem for AI data rights management.

Conversely, a Salesforce win might cement fair use as a shield for large-scale training, leaving creators sidelined from the digital economy built on their words.

This debate isn’t confined to literature. Similar disputes are unfolding across industries, including film and design as seen in Disney & Universal vs. Midjourney: Inside the AI Copyright Battle That Could Rewrite Hollywood Law.

Together, these cases mark a global turning point for how law defines creativity in the age of algorithms.

Final Thought

The Tanzer v. Salesforce case goes beyond legal arguments, it’s part of a larger conversation about what creativity means in the age of machines.
If the authors win, it could mark the start of a new era where writers, artists, and creators are finally recognized and compensated for the value their work brings to artificial intelligence.

If Salesforce prevails, it may set a precedent that blurs the line between inspiration and imitation, raising uncomfortable questions about who truly owns creative expression in a digital world.

Whatever the outcome, the decision will ripple far beyond Silicon Valley, shaping how society balances innovation, ownership, and the human voice within AI’s expanding reach.

People Also Ask (PAA)

What is the Salesforce AI copyright lawsuit about?
The case involves authors accusing Salesforce of using their copyrighted books without permission to train its xGen AI model. They claim this violates the U.S. Copyright Act and undermines creative ownership in the age of artificial intelligence.

Why are authors suing Salesforce?
Writers Molly Tanzer and Jennifer Gilmore filed a class action alleging that Salesforce’s AI learned from pirated or unlicensed works. Their lawsuit seeks damages and stronger legal protection for creative content used in AI training.

Is it legal to use copyrighted books to train AI models?
The legality depends on fair use — a doctrine that allows limited use of copyrighted material for transformative purposes. Courts must now decide whether teaching AI to generate new text counts as transformation or infringement.

What could happen if Salesforce loses the lawsuit?
If the authors prevail, Salesforce may face major financial penalties and be forced to license copyrighted data. The decision could also set a national precedent requiring all AI developers to pay for the creative works they use.

How could this case impact future AI laws?
A ruling against Salesforce could shape how lawmakers regulate data transparency and copyright licensing in AI development. It may redefine fair use, forcing companies to rethink how they train large language models.

When Elon Musk’s Grok app unveiled its Imagine feature this summer, most of the headlines focused on the playful (and controversial) “Spicy” mode, which allows looser, less filtered outputs. But beneath the hype, a smaller update may prove more consequential: the Access to Speech feature.

This new tool lets users generate short AI-animated videos simply by speaking prompts aloud. On the surface, it looks like a usability perk. For lawyers, regulators, and anyone watching the fast-moving legal tech space, it raises much bigger questions about accessibility, authorship, liability, and data rights.

What Grok Imagine Does

Grok Imagine, launched in mid-2025, is Musk’s answer to the growing field of AI video tools. Think of it as a revival of Vine, only instead of filming your own clips, you provide a prompt typed or spoken and the system generates a 6–15 second animated video with sound.

The tool offers four creative modes:

• Normal (mainstream safe use)

• Fun (lighthearted, playful results)

• Custom (tailored outputs)

• Spicy (minimal content restrictions, and already a lightning rod for deepfake concerns)

While casual users play with Spicy, lawyers should pay close attention to the quieter voice-input feature.

Accessibility Meets Compliance

Allowing voice input makes Grok Imagine more inclusive, particularly for users with limited mobility or difficulty typing. In theory, this could support compliance with accessibility laws such as:

• The Americans with Disabilities Act (ADA) in the U.S.

• The Equality Act 2010 in the UK.

But legal tech experts know accessibility is a double-edged sword. Once features like voice input become widely available, courts and regulators may begin asking: is this now a standard? If so, companies that lack comparable accessibility options could face pressure or even liability for failing to provide them.

Intellectual Property: Who Owns a Spoken Prompt?

The more complex legal issue comes with authorship. If a user speaks an idea into Grok Imagine and the AI generates a video, who owns the output?

• In the U.S., the Copyright Office has consistently refused to grant copyright to works “not created by a human author.”

• Yet if a human provides the spoken idea, is there a case for shared authorship?

• In commercial contexts say, advertising generated from voice prompts this grey zone could become a litigation risk.

As AI video tools expand, expect intellectual property lawyers to confront more disputes over ownership of “voice-to-video” works.

Deepfakes and Liability: The Spicy Factor

The legal risks grow when Access to Speech intersects with Spicy mode. A user can simply say “make a video of [celebrity name] in X scenario” and produce a deepfake-style clip within seconds.

That raises at least three major concerns:

1. Right of publicity – In the U.S., celebrities and private individuals alike can sue for unauthorized commercial use of their likeness.

2. Defamation – Harmful or false portrayals could expose both users and platforms to claims.

3. Privacy – EU and UK law offers strong protections against misuse of personal images and reputations.

So where does liability land? With the platform (xAI)? With the user? Or both? Courts are likely to wrestle with this as AI video tools spread.

Voice Data as Biometric Information

Voice prompts mean Grok is collecting voice recordings, which may be treated as biometric data in some jurisdictions. That triggers strict compliance regimes:

• GDPR (EU/UK) – classifies voice as personal data, requiring clear consent and purpose limitations.

• California CCPA/CPRA – expands consumer rights over stored data.

• Illinois BIPA – one of the toughest U.S. biometric privacy laws, with steep penalties for mishandling voice data.

For lawyers advising tech companies, the message is simple: Access to Speech isn’t just a convenience. It’s a data collection pipeline that could carry significant regulatory risk if not handled properly.

AI Regulation: A Moving Target

Finally, there’s the bigger picture: how regulators will treat generative AI in the years ahead.

• The EU AI Act, due to roll out gradually from 2025, will classify AI systems by risk and impose obligations around transparency, labeling, and safety. Grok Imagine’s voice-to-video system could fall under “general-purpose” or “high-risk” categories, depending on usage.

• In the U.S., while there’s no federal AI law yet, states like California, New York, and Illinois are pushing their own AI and deepfake legislation.

Access to Speech may look harmless, but as regulators move faster, it will likely be swept into broader compliance frameworks.

The Access to Speech feature highlights how even small usability upgrades in AI tools can carry wide legal implications. Lawyers and compliance officers should be alert to:

1. Accessibility – Could soon be treated as a legal obligation.

2. Copyright and authorship – Voice-to-video works sit in a legal grey zone.

3. Deepfake misuse – Spicy mode + voice prompts = higher litigation risk.

4. Biometric privacy – Voice data may trigger GDPR, CCPA, or BIPA duties.

5. Emerging regulation – EU AI Act and state laws will reshape the landscape.

For most Grok users, Access to Speech feels like a fun shortcut. For the legal community, it’s a case study in how rapid innovation collides with slow-moving law.

Voice-to-video tools touch on nearly every corner of modern legal practice - accessibility, IP, media law, privacy, and regulation.

In short, talking to AI is easy. Working out who owns, protects, and regulates what comes next is anything but.

People Also Ask (PAA)

What is Grok Imagine?
It’s an AI feature in Elon Musk’s Grok app that generates short animated videos (6–15 seconds) with sound, based on text or voice prompts.

What does “Access to Speech” mean in Grok?
It refers to the ability to speak prompts into the app instead of typing them, making video generation faster and more accessible.

Why does Access to Speech raise legal issues?
Because it touches on accessibility law, copyright ownership, data privacy (voice as biometric data), and liability for misuse (e.g., deepfakes).

Is Access to Speech available to all Grok users?
It began rolling out to premium subscribers on iOS and is gradually expanding to more users.

What laws could apply to Grok’s voice feature?
Accessibility laws (ADA, Equality Act 2010), copyright law, biometric privacy laws (GDPR, BIPA, CCPA), and upcoming AI-specific regulations like the EU AI Act.

]]>